We've gone full circle - Flat to fully routed to MPLS/VPLS over a routed
network back to flat.  You hit a scaling issue with routed networks as you
hit 10G and above, especially if you aren't using Mikrotik or other  low
cost routing.  Real carrier grade switching is a lot lower cost, lower
power, and much easier to manage.

Every customer has their own dedicated circuit (SVLAN.CVLAN).  The
corresponding interface on the BNG is dynamically created for the
subscriber with attributes out of radius.   Something like this isn't the
right answer at 100 customers but you should consider it or something like
it once you go north of a few k subs.

On Fri, Jun 18, 2021 at 10:32 AM Jan-GAMs <j.vank...@grnacres.net> wrote:

> We're under 100 subs, and static routing has been easy to monitor via the
> UISP.  Every CPE is displayed and easy to login to.  Any units on DHCP is a
> total PITA and I'd prefer to shoot the guy that started doing that as we
> can't find the user nor login to fix them, it's a truck roll which consumes
> hours instead of 5 minutes.  I would like to know more about setting up
> networks and finding the users once DHCP is deployed, since this is what
> our IT group is doing, drives me crazy.  I like to run a pro-active service
> dept instead of waiting for complaints.
>
> Well we could replace the switches with routers but won't this reduce the
> total traffic available?  And once the traffic passes through several
> towers it gets reduced to not much available, more so than wide open links
> in bridge mode?
> On 6/18/21 8:16 AM, Sam Lambie wrote:
>
> We had a flat network for a few years with the same setup as you in terms
> of network. Once the network grew to a certain size, broadcast storms would
> roll through often and it was almost impossible to track down the culprit
> without unplugging the gear and waiting for it to die down. We then
> switched to VLANs and that helped immensely, but was hard to manage as most
> everything was static routes and we had to remember where everything was
> routed.
> Finally, we moved to routers at each tower, Natting the whole network and
> doing a whole slew of other things at the core side and so far for the past
> few years, it has been nice.
>
> On Fri, Jun 18, 2021 at 9:04 AM Jan-GAMs <j.vank...@grnacres.net> wrote:
>
>> I think this is beyond our present capability.  We have an edgerouter X
>> where the network meets the internet and that's it.  There is only one
>> OSPF, it's just one path with no other routes.  We have a switch at every
>> tower that powers the APs and clients(CPE) that connect to APs.  We use
>> UISP to monitor the network remotely.  Each CPE radio is a router but all
>> are in "bridge" mode and we have different brands of routers inside the
>> customer homes, non-ubnt devices are using dhcp.  We use one VLAN for
>> management.  All customers are set to 20MBps for traffic control.
>>
>> I couldn't find the guilty radio if there was one and the traffic being
>> shown at the final uplink to the outside world would only pass about
>> 0.1kbps using the built-in speedtest between it and the next closest link
>> but the traffic monitor was showing peaks of about 6Mbps for total
>> traffic.  I found nothing that could prove the traffic was real.
>>
>>   There doesn't seem to be enough functions available in the CPEs to
>> actively prevent this problem from happening again.  I'm not sure what you
>> mean by "multicast"?  It makes sense to figure out a way to squelch it.
>> On 6/18/21 7:15 AM, Adam Moffett wrote:
>>
>> This is plausible.  I think ubnt sends broadcast traffic at MCS0.  Not
>> sure how it handles multicast.  If everyone was in the same layer2 domain a
>> heavy broadcast traffic could affect the whole system.  Maybe the customer
>> moving 6-10mbps was malfunctioning and broadcasting something.
>>
>> In general it's safe to block all multicast and only allow it where you
>> need to make OSPF connections.  Broadcast can be limited to 10kbps per
>> customer with no issue.  The only broadcast they need to function is an ARP
>> for their default gateway and a DHCP discover.  After initial discovery the
>> DHCP traffic switches to unicast.  Not sure what tools ubnt gives you for
>> filtering that, but ideally you'd block multicast and limit broadcast at
>> every CPE.
>>
>>
>> On 6/18/2021 9:33 AM, Daniel White wrote:
>>
>> Sounds like a broadcast storm to me.  What is the topology of your
>> network?  Routers at each tower, VLANs, etc.?
>>
>> Are you filtering multicast and broadcast traffic at the CPE/customer
>> premises?
>>
>> [image: photograph]
>> Daniel White
>> Co-Founder
>> phone: +1 (702) 470-2770
>> direct: +1 (702) 470-2766
>>
>> Jan-GAMs <j.vank...@grnacres.net>
>> June 17, 2021 at 23:47
>> We had a strange outage on one of our networks yesterday.  At first we
>> thought it was one customer.  The symptom was very low to non-existent
>> internet traffic.  The complaint was my internet is not working!
>>
>> Upon testing I found that the complaining customer had for a speed test
>> about 0.14kbps for a speed to it's AP.  So I went to their AP and tested
>> the speed back at them, it was about the same unusually slow speed.  Then I
>> tested that AP to another AP and that speed was about the same slow speed.
>> So then I tested another customer and another and then ended up testing
>> just about everyone in the whole network.  Everyone was operating at an
>> unusually slow speedtest to any other device of about 0.1kbps to 0kbps.
>> The whole network was down and yet the UISP was indicating everyone was up
>> and operating with even some traffic in the 6-10 Mbps range which I'm sure
>> was fake traffic as none of the devices tested would pass anything above a
>> few kbps.
>>
>> A reboot of every device resolved the issue.
>>
>> Our gear is Ubiquiti and I'm wondering has anyone else using Ubiquiti
>> been experiencing anything like what I just described? Is there a known
>> cause?
>>
>>
>>
>>
>>
>> --
>> AF mailing list
>> AF@af.afmug.com
>> http://af.afmug.com/mailman/listinfo/af_af.afmug.com
>>
>
>
> --
> --
> *Sam Lambie*
> Taosnet Wireless Tech.
> 575-758-7598 Office
> www.Taosnet.com <http://www.newmex.com>
>
> --
> AF mailing list
> AF@af.afmug.com
> http://af.afmug.com/mailman/listinfo/af_af.afmug.com
>


-- 

Carl Peterson

*PORT NETWORKS*

401 E Pratt St, Ste 2553

Baltimore, MD 21202

(410) 637-3707
-- 
AF mailing list
AF@af.afmug.com
http://af.afmug.com/mailman/listinfo/af_af.afmug.com

Reply via email to